Microsoft launches in-house AI models, cuts costs up to 89%

Microsoft has introduced two new in-house AI models — MAI-Image-2.5-Pro and MAI-Voice-2-Flash — now available in public preview. The company claims these models can reduce inference costs by up to 89% compared to equivalent models from OpenAI, such as DALL-E and Whisper. The announcement, made by Microsoft's Superintelligence team, reinforces the company's strategy to reduce dependence on external suppliers and build its own AI capabilities.

From a technical standpoint, MAI-Image-2.5-Pro is the highest-fidelity image generator Microsoft has ever released, delivering superior detail, color accuracy, and consistency. MAI-Voice-2-Flash is designed for high-volume enterprise workloads, with reduced latency and improved contextual understanding. Both models were trained on Microsoft's own infrastructure and optimized to run on Azure, which explains part of the cost advantage.

For the technology and business market, this move represents a concrete opportunity to reduce operational expenses. Companies using voice models for call centers or image generation for marketing campaigns can migrate to the new models and achieve significant scale gains without compromising quality. Autonomy from OpenAI also reduces vendor lock-in risks and price fluctuations.

From the growth marketing perspective at 10Dobro Prod, the reduction in cost per inference is a multiplier for experimentation. With cheaper models, marketing teams can test many more visual and audio creative variations, automate voice responses in sales funnels, and personalize content in real time without breaking the budget. This accelerates learning cycles and improves conversion and retention rates. Specific audiovisual applications, such as generating images for dynamic ads, benefit directly, but the real impact lies in being able to scale AI-driven automation at an accessible cost.